Diagnostic accuracy of quantitative flow ratio (QFR) and vessel fractional flow reserve (vFFR) estimated retrospectively by conventional radiation saving X-ray angiography

Abstract

Background: Angiography derived FFR reveals good performance in assessing intermediate coronary stenosis. However, its performance under contemporary low X-ray frame and pulse rate settings is unknown. We aim to validate the feasibility and performance of quantitative flow ratio (QFR) and vessel fractional flow reserve (vFFR) under such angiograms.

Methods: This was an observational, retrospective, single center cohort study. 134 vessels in 102 patients, with angiograms acquired under 7.5fps and 7pps mode, were enrolled. QFR (fQFR and cQFR) and vFFR were validated with FFR as the gold standard. A conventional manual and a newly developed algorithmic exclusion method (M and A group) were both evaluated for identification of poor-quality angiograms.

Results: Good agreement between QFR/vFFR and FFR were observed in both M and A group, except for vFFR in the M group. The correlation coefficients between fQFR/cQFR/vFFR and FFR were 0.6242, 0.5888, 0.4089 in the M group, with r_vFFR significantly lower than r_fQFR (p = 0.0303), and 0.7055, 0.6793, 0.5664 in the A group, respectively. AUCs of detecting lesions with FFR ≤ 0.80 were 0.852 (95% CI 0.722-0.913), 0.858 (95% CI 0.778-0.917), 0.682 (95% CI 0.586-0.768), for fQFR/cQFR/vFFR in the M group, while vFFR performed poorer than fQFR (p = 0.0063) and cQFR (p = 0.0054). AUCs were 0.898 (95% CI 0.811-0.945), 0.892 (95% CI 0.803-0.949), 0.843 (95% CI 0.746-0.914) for fQFR/cQFR/vFFR in the A group. AUC_vFFR was significantly higher in the A group than that in the M group (p = 0.0399).

Conclusions: QFR/vFFR assessment is feasible under 7.5fps and 7pps angiography, where cQFR showed no advantage compared to fQFR. Our newly developed algorithmic exclusion method could be a better method of selecting angiograms with adequate quality for angiography derived FFR assessment.

Keywords: Angiography; Fractional flow reserve; QFR; vFFR.

Fig. 1

Typical sample images of different grades of each variable enrolled in the algorithmic excluding method (interrogated vessels are labeled at the right-up corner, image quality not acceptable is not shown due to lack of sample). Specific definition of each grade could be found in the supplements

Fig. 2

Bland–Altman plots of intra and inter observer reliability. For intra observer reliability, the mean difference between the 2 groups of measurements by observer 1 was 0.016 ± 0.060 (p = 0.066) for fQFR, 0.009 ± 0.053 (p = 0.230) for cQFR, and 0.008 ± 0.040 (p = 0.175) for vFFR, respectively. Inter-observer difference was also non-significant, with 0.001 ± 0.036 (p = 0.847) for fQFR, − 0.001 ± 0.049 (p = 0.910) for cQFR, and − 0.005 ± 0.037 (p = 0.393) for vFFR

Fig. 3

Study flowchart

Fig. 4

Agreement and correlation between QFR/vFFR and FFR, and ROC curves of identifying FFR ≤ 0.80. For both a and b panel, upper part: Bland–Altman plots of differences against the means; lower part: correlation between QFR/vFFR and FFR. For c and d panel, ROC curves of QFR/vFFR and their comparison within same group

Fig. 5

Vessel contour detection with necessary manual modifications. Same vessel (an intermediate ramus) was assessed by both QFR and vFFR software package. A Initial frame without any projections. A1 Semi-automatic detection of vessel contours with necessary manual modifications by QFR software package. A2 3D reconstruction of the vessel by QFR software package. B1 semi-automatic detection of vessel contours with necessary manual modifications by vFFR software package, the contours are not so smooth and jagged. B2 Same vessel contours after being modified with “Hard correction” tool provided by the software package, the vessel contours become smoother. B3 3D reconstruction of the vessel by vFFR software package, the reconstructed vessel wall is not as smooth as that in QFR software